[jira] [Comment Edited] (TINKERPOP-1849) Provide a way to fold() with an index

["Marko A. Rodriguez (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/TINKERPOP-1849?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16282342#comment-16282342 ] 

Marko A. Rodriguez edited comment on TINKERPOP-1849 at 12/7/17 7:06 PM:
------------------------------------------------------------------------

After bragging to [~dkuppitz] about my solution, he said: "Does it work on GraphComputer?" I immediately thought: "no" and tested it and it doesn't. Why? {{FoldStep}} is a {{ReducingBarrierStep}} and thus, can only work with associative and commutative operations. {{addWithIndex}} is not. It requires a serial ordering during "folding" in order to get an index. Thus, this raises two solutions:

Do we provide a "flag" on {{Operators}} to say whether they are associative and commutative?
  1. If so, then if it is not, then all reduction must happen on the master traversal.
  2. Or, simply through a {{ComputerVerificationException}} is such operators are used.

The more general problem would be provided lambdas that are NOT commutative nor associative. How is Gremlin to know?


was (Author: okram):
After bagging to [~dkuppitz] about my solution, he said: "Does it work on GraphComputer?" I immediately thought: "no" and tested it and it doesn't. Why? {{FoldStep}} is a {{ReducingBarrierStep}} and thus, can only work with associative and commutative operations. {{addWithIndex}} is not. It requires a serial ordering during "folding" in order to get an index. Thus, this raises two solutions:

Do we provide a "flag" on {{Operators}} to say whether they are associative and commutative?
  1. If so, then if it is not, then all reduction must happen on the master traversal.
  2. Or, simply through a {{ComputerVerificationException}} is such operators are used.

The more general problem would be provided lambdas that are NOT commutative nor associative. How is Gremlin to know?

> Provide a way to fold() with an index
> -------------------------------------
>
>                 Key: TINKERPOP-1849
>                 URL: https://issues.apache.org/jira/browse/TINKERPOP-1849
>             Project: TinkerPop
>          Issue Type: Improvement
>          Components: process
>    Affects Versions: 3.3.0
>            Reporter: stephen mallette
>
> In Groovy you can call {{withIndex()}} to generate output like this:
> {code}
> gremlin> g.V().fold().next().withIndex()
> ==>[v[1],0]
> ==>[v[2],1]
> ==>[v[3],2]
> ==>[v[4],3]
> ==>[v[5],4]
> ==>[v[6],5]
> {code}
> We can currently simulate this with Gremlin via:
> {code}
> gremlin> g.V().project("a","b").by().by(select("x").count(local)).store("x").map(union(select('a'), select('b')).fold()).fold().next()
> ==>[v[1],0]
> ==>[v[2],1]
> ==>[v[3],2]
> ==>[v[4],3]
> ==>[v[5],4]
> ==>[v[6],5]
> {code}
> but it's not easy to follow, efficient, or potentially foolproof. Perhaps we can add an option to {{fold()}} that would take an enum of "fold operators".



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)